# Architecture Overview

**Hybrid AI + Edge SCADA for Small Water Systems**

## Core Principles

- **Local First**: As much processing as possible happens on the edge device.
- **Low Egress**: Healthy state = highly compressed summaries. Only alerts carry rich context.
- **Hybrid Intelligence**: Small models on edge for real-time decisions + larger models for complex reasoning and onboarding.
- **Monitored Objects** as the central abstraction.
- **Multi-Tenancy** with clear RBAC (Owner, Operator, User, Field Tech).

## High-Level Components

1. **Edge Device** (Raspberry Pi + Cellular)
   - Local real-time signal processing
   - Local web UI for setup (hotspot mode)
   - Camera support for analog sensors
   - eSIM / cellular connectivity

2. **Cloud / Server Layer**
   - Multi-tenant web application
   - Dashboard (Basic + Advanced modes)
   - Alert management and history
   - Operator multi-system view
   - Strong model for complex tasks (onboarding interview, anomaly detection, recommendations)

3. **AI Model Strategy**
   - Small specialized models on edge
   - 9B-class model for personality/continuity and triage decisions
   - Larger models (30B+) for complex reasoning and onboarding

## Data Flow

- Normal operation: Edge summarizes and uploads compressed data every 30-60 min
- Alert condition: Edge sends rich payload immediately + triggers notifications

**See**: `specs/` folder for detailed specifications.